#version 450 core
#extension GL_KHR_memory_scope_semantics : enable
#extension GL_NV_cooperative_vector : enable
#extension GL_EXT_shader_explicit_arithmetic_types : enable
#extension GL_EXT_buffer_reference : enable
#extension GL_EXT_nonuniform_qualifier : enable

layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;

layout(set = 0, binding = 0) buffer MatrixBuf {
    float16_t matrixData[];
} matrixBuf;

layout(set = 0, binding = 0) buffer MatrixBuf2 {
    int8_t matrixData[];
} matrixBuf2;

layout(constant_id = 0) const int matrixLayout = gl_CooperativeVectorMatrixLayoutColumnMajorNV;

void main()
{
    coopvecNV<float, 5> v = coopvecNV<float, 5>(0.0);

    vec3 x = vec3(1.0), y = vec3(2.0);
    coopvecNV<float, 5> v2 = coopvecNV<float, 5>(x, y);

    const vec3 x2 = vec3(1.0), y2 = vec3(2.0);
    coopvecNV<float, 5> v3 = coopvecNV<float, 5>(x2, y2);

    vec3 x3 = vec3(4, 5, 6), y3 = vec3(7, 8, 9);

    coopvecNV<float, 5> v4 = coopvecNV<float, 5>(x3.zyx, y3.zzy);

    coopvecNV<float16_t, 5> v5 = coopvecNV<float16_t, 5>(v4);

    coopvecNV<float, 5> v6;

    v6 = v + v2;
    v6 = v - v2;
    v6 = v * v2;
    v6 = v / v2;

    v6 += v;
    v6 -= v;
    v6 *= v;
    v6 /= v;

    float f;
    v6 *= f;
    v6 *= 5.0;

    int len = v6.length();

    v6[0] = f;
    v6[4] = 5.0;

    f += (v += v)[4];
    v6 = v;

    coopvecNV<float, 20> v7;
    coopvecNV<float, 100> v8;
    coopVecMatMulNV(v8, v7, gl_ComponentTypeFloat16NV, matrixBuf.matrixData, 11, gl_ComponentTypeFloat16NV, 100, 20, gl_CooperativeVectorMatrixLayoutRowMajorNV, false, 300);
    coopvecNV<int8_t, 20> v9;
    coopvecNV<int32_t, 100> v10;
    uint32_t matrixOffset = 11;
    coopVecMatMulNV(v10, v9, gl_ComponentTypeSignedInt8NV, matrixBuf2.matrixData, matrixOffset, gl_ComponentTypeSignedInt8NV, 100, 20, gl_CooperativeVectorMatrixLayoutColumnMajorNV, true, 300);

    coopvecNV<float16_t, 2> f162 = coopvecNV<float16_t, 2>(v7.length());

    float16_t f16;
    coopvecNV<float16_t, 1> f161 = coopvecNV<float16_t, 1>(f16);

    coopvecNV<float16_t, 7> v11, v12, v13;

    v11 = max(v11, v12);
    v11 = min(v11, v12);
    v11 = step(v11, v12);
    v11 = clamp(v11, v12, v13);
    v11 = exp(v11);
    v11 = log(v11);
    v11 = tanh(v11);
    v11 = atan(v11);
    v11 = fma(v11, v12, v13);

    coopVecMatMulAddNV(v8, v7, gl_ComponentTypeFloat16NV, matrixBuf.matrixData, 11, gl_ComponentTypeFloat16NV, matrixBuf.matrixData, 50, gl_ComponentTypeFloat16NV, 100, 20, gl_CooperativeVectorMatrixLayoutRowMajorNV, false, 300);
    uint32_t biasOffset = 11;
    coopVecMatMulAddNV(v10, v9, gl_ComponentTypeSignedInt8NV, matrixBuf2.matrixData, matrixOffset, gl_ComponentTypeSignedInt8NV, matrixBuf2.matrixData, biasOffset, gl_ComponentTypeSignedInt8NV, 100, 20, gl_CooperativeVectorMatrixLayoutColumnMajorNV, true, 300);

    coopvecNV<uint32_t, 20> v14;
    coopVecMatMulNV(v10, v14, gl_ComponentTypeSignedInt8PackedNV, matrixBuf2.matrixData, matrixOffset, gl_ComponentTypeSignedInt8NV, 100, 20, matrixLayout, true, 300);

    f += (v += v)[len];
}

coopvecNV<float16_t, 1> foo(coopvecNV<float16_t, 1> x) { return x; }
coopvecNV<float16_t, 2> foo(coopvecNV<float16_t, 2> x) { return x; }